Fix PostgreSQL compound key discovery

[strk]

Closes GH-56420

[strk]

Tests for relation discovery seem to exist but are based on referenced keys being first attributes of the referenced table. I'm going to update that one.

To run:

QGIS_PREFIX_PATH=${builddir}/output PYTHONPATH=${builddir}/output/python:$PYTHONPATH python tests/src/python/test_provider_postgres.py TestPyQgsPostgresProvider.testValidLayerDiscoverRelationsComposite

[github-actions]

🪟 Windows builds ready!

Windows builds of this PR are available for testing here. Debug symbols for this build are available here.

(Built from commit 847ef3e)

[strk]

I've pushed a change in existing testcase to basically list the referenced columns in the inverse order in which they were defined. This triggers the bug reported in GH-56420

To show the CI catching the bug I've currently also pushed a revert of the fix, to expect a failure. Once the failure is shown, I'll drop the revert and leave the fix in place to see how that fixes the problem.

[strk]

Bug confirmed by CI: https://github.com/qgis/QGIS/actions/runs/9446127636/job/26016503531?pr=57653#step:13:2257
Now I'll force-push a version of the branch with the fix in place

[github-actions]

Tests failed for Qt 5

One or more tests failed using the build from commit 847ef3e

layout_export_markerline_masked_geometry

layout_export_markerline_masked_geometry

Test failed at test_markerline_masked at tests/src/python/test_selective_masking.py:1376

Rendered image did not match tests/testdata/control_images/selective_masking/layout_export_markerline_masked/layout_export_markerline_masked.png (found 124 pixels different)

The full test report (included comparison of rendered vs expected images) can be found here.

Further documentation on the QGIS test infrastructure can be found in the Developer's Guide.